Personal Note 03 · AI, sky and uncertainty
A Romantic Bet with the Sky
Everyone said AI should improve efficiency. I wanted it to anticipate something beautifully unpredictable.
Toronto often gave me blue hour. Singapore became known to me through pink sunsets. Comparing skies with friends in New York, Toronto, the UK and Harbin, I began to wonder whether latitude had something to do with the colours we remembered.
I asked ChatGPT about the sky
I wanted to know how a pink sky might be predicted. ChatGPT could not answer whether tomorrow evening would turn pink, but it gave me directions to investigate: cloud cover, humidity, particles in the air, the angle of the setting sun.
People spoke about AI as an efficiency machine that needed to become useful and profitable quickly. I wanted it to help me anticipate unpredictable romance.
A model I imagined, and a smaller thing I made
I imagined collecting years of sunset photographs and weather records, then training a model to learn the conditions of a pink sky. I did not build that model. I matched only a few photographs with weather information and made something much smaller.
The working prototype retrieved live Singapore weather and translated selected conditions into an experimental probability. It was not a scientific forecast. It was the first computational form I could give to the question.
More than two hours for a first piece of code
I was not a programmer. Even a first running script took me more than two hours to assemble with GPT guiding me through it. The code was simple, but the experience mattered: an idea that had existed as a conversation about skies became something I could open, refresh and test.
I added a rotating literary reference because a probability alone felt too dry. I did not want another weather dashboard. I wanted something closer to a letter addressed to the sky.
What should a probability ask of me?
A number can tell me how confident a system is. It cannot decide whether a sunset will matter to me. Pink Sky Predictor made me interested in the space between those things.
If the probability is high, perhaps I will remember to look. If it is low, I may still choose to wait. The calculation does not remove uncertainty. It gives uncertainty a time and place in the day.
A reason to look up
The project did not prove that AI can predict romance. It made me realise that computation does not always have to see on our behalf.
Sometimes its most humane role may be smaller: to interrupt the efficient day, create a little expectation and return the final act of noticing to us.
Tomorrow may not have a pink sky. I still like having a reason to look.